How to Make the ESPN AI Video (Put Your Name on the Broadcast)

What You're Trying to Make

You know the shot: the house lights drop, a spotlight hits a single player, the crowd roars, and a broadcast graphic sweeps on-screen with the starting lineup — name, number, position, stat line. This is that moment, except the player is you, generated from one photo. The lower-third animates on like a real motion-graphics package, then locks, while you stand there under the arena lights, quietly alive.

This guide walks through how to make one: what photo to use, how to render the graphic so the text is crisp, the exact prompt that makes the lower-third build on instead of sitting there static, which model to run it on, and how to post it. By the end you'll have a repeatable, generation-ready workflow.

Fastest way — the Starting Lineup template does all of this from one face photo: it renders your personalized name-and-stats broadcast frame, animates the lower-third on, and hands back a 10-second prime-time intro in a few minutes — 1 credit, no prompt writing. Want to roll your own and pick your own stat line? The full method's below. ↓

Where This Came From

In June 2026, ESPN aired AI-generated "living portraits" of NBA legends during the Finals — iconic still photos of players like Tony Parker subtly animated to move as commercial-break bumpers. The internet roasted them as "AI slop" (the re-rendered faces drifted so far they barely looked like the players), and the network pulled them after Game 1.

The creator flip writes itself. Instead of a network animating a legend — and getting it wrong by over-moving the face — you put yourself on the broadcast, on purpose, as a personalized player introduction. It's the named-chyron member of the broadcast fan-cam family that's been running all year: the baseball fan cam, the courtside NBA cut, the F1 paddock, the World Cup fan cam. The difference: those catch you in the crowd. This one gives you a name, a number, and a stat line — the starting-lineup graphic, animating on, all yours.

The lesson from ESPN's misfire is baked into the recipe below: the face barely moves. Over-animation is exactly what got the real portraits mocked. The motion belongs to the graphic, not the person.

The Fastest Way — Use the Starting Lineup Template on Starrd

The Starting Lineup template is live in the Starrd library. It packages every step in this guide — the finished broadcast frame with your personalized name and stats, the animated lower-third that builds on, the identity lock, the de-brand — into a single upload.

1. Pick a clear face photo. One person, front or three-quarter view, eyes open, decent lighting.
2. Open the Starting Lineup template in the Starrd app or web library.
3. Upload the photo and tap generate. The template renders your personalized "starting lineup" broadcast frame with an image model strong at text, then animates the lower-third on and generates a 10-second prime-time intro on Seedance 2.0 — clean open, the graphic builds in, locks, you settle into the camera.

One credit, a few minutes. No prompt writing, no model picking. Add your name, number, position, stats, or a whole different sport in director's notes and it rebuilds the graphic around them.

The rest of this guide is for people who want to roll their own — pick your own stat line, swap the sport, or run it on a different model.

Or, Build It Yourself — What You Need

Three things:

1. A clear face photo of the subject. Front or three-quarter view, eyes open, good lighting. One person only.
2. An image model strong at dense on-screen text (for the finished frame with the lower-third) plus a video model that accepts a reference image (to animate it). GPT Image 2 for the frame, Seedance 2.0 for the animation, is the crispest combo.
3. A platform to post on. TikTok, Reels, and X are where broadcast formats go viral.

You don't need editing software, After Effects, or a jersey.

Step 1 — Pick Your Reference Photo

The photo you feed the model is the face that ends up on the broadcast. Choose well and you save wasted generations.

Use:

• A clear, well-lit photo of one person
• Front-facing or three-quarter angle
• Eyes open, neutral or faintly confident "game face"
• Natural lighting, minimal filtering

Avoid:

• Group photos (the model gets confused about who's who)
• Sunglasses or anything covering the face
• Low-resolution or motion-blurred shots
• A recognizable celebrity face (the model fights identity lock and viewers instantly clock it)

Step 2 — Decide Your Name, Number, Position, and Stat Line

Lock these before you render anything — they're the words the graphic has to spell correctly.

• Name: use a short surname in the chyron. Long full names shrink the type and invite garbling. REED reads; CHRISTOPHER REEDINGTON JR. does not.
• Number: any jersey number. It appears on the kit and in the graphic's number block.
• Position + stat line: pick a sport and its three headline stats. Basketball: GUARD · PTS 28 · REB 6 · AST 11. Football: QUARTERBACK · YDS · TD · QBR. Soccer: STRIKER · G · A · SOG. Keep the numbers plausible and short.

Three stat modules is the sweet spot — enough to read as a real broadcast card, few enough that the model renders every character cleanly.

Step 3 — Write the Prompt

This is a two-part recipe, and that split is the whole trick.

Part A — the finished "end-state" frame (image model). Render one still: you at broadcast distance (waist-up, face well under 40% of the frame, crowd behind) in a dimmed arena with intro spotlight beams, haze, and LED boards — with the full lower-third already composited in. Let the image model, which is far better at crisp text than a video model, bake the graphic perfectly.

Part B — animate it on (video model). Feed that frame to the video model as the reference, and describe the lower-third as a timed 2D graphics overlay that builds on and then locks to match the reference. This is what makes it move instead of sitting there.

Treat @Image1 as the FINISHED broadcast player-introduction frame — it defines the exact end-state design, layout, and TEXT of the lower-third (a de-branded "starting lineup" stat card, lower-left: royal-blue number block "11", surname "REED" in condensed white caps, "GUARD" beneath, and "PTS 28 · REB 6 · AST 11"). The clip animates INTO this state. The lower-third is a flat 2D broadcast graphics overlay in screen space — it animates independent of the camera, then locks to match @Image1.Single broadcast camera, essentially locked off with only the faintest drift. Hold broadcast distance the whole time (waist-up, face under 40% of frame) — never a tight close-up.[0-1s] Opens CLEAN — the player is hero-lit and subtly alive (breathing, a slight settle, eyes toward the lens), NO lower-third yet.
[1-3s] The lower-third ANIMATES ON in sequence like a live-TV motion-graphics package: the number block slides up from the lower edge and snaps in; the name panel wipes open to the right, unfurling "REED"; a bright light-sweep glints across the glass; the position line fades up; the stat modules pop on one by one, left to right; a thin accent line draws across the base. Each element lands with a broadcast whoosh.
[3-6s] Fully assembled, matching @Image1 exactly — the graphic LOCKS and holds perfectly static and crisp (no warping, no value changes). The player gives a slight head turn toward the lens and a confident chin-lift, one slow blink.
[6-10s] The player holds steady eye contact, calm intensity; camera flashes twinkle in the crowd bokeh. The locked lower-third stays crisp to the end.Audio: a hyped arena PA intro — "AND NOW... YOUR STARTING POINT GUARD... NUMBER ELEVEN... REEED!" — the crowd erupts on the name, motion-graphics whoosh as the graphic builds, a driving intro sting, then crowd roar. No real team or network names spoken.Premium prime-time broadcast realism, broadcast lens compression, natural skin texture, NOT a beauty filter. Avoid: the graphic appearing fully-formed at 0s, warping the locked text or changing its numbers, a tight close-up, the face filling the frame, a wide smile forming, any real network or team logo, a "LIVE" tag.

The non-negotiable elements:

• "Treat @Image1 as the FINISHED end-state... animates INTO this state" — this is what lets the graphic build on while still matching a crisp, pre-rendered design.
• "Opens CLEAN... NO lower-third yet" — without it, the model renders the graphic fully-formed from frame one and you lose the whole build-on.
• The sequence of mograph verbs — "slides up," "wipes open," "light-sweep glints," "pop on," "draws across." Broadcast packages animate element by element; naming each beat is what produces the build.
• "LOCKS and holds perfectly static" — after it assembles, the graphic must freeze. A chyron that keeps drifting or re-rendering its numbers reads as fake instantly.
• The face barely moves — a breath, one blink, a slight settle. This is the ESPN lesson: over-animating the face is the tell.

Step 4 — Pick a Model

• Seedance 2.0 — what Starrd runs on. Best-in-class prompt adherence, which is why it can open clean and animate the lower-third on cue, and it renders legible text plus the PA/crowd audio. The safest pick.
• GPT Image 2 — for Part A, the still end-state frame. It's the strongest at dense, correctly-spelled on-screen text, so the baked lower-third comes out crisp.
• Kling 3.0 / Runway Gen-4 / Veo 3.1 — solid at animating the person, but less reliable at building a text graphic on. Usable if you keep the graphic simple.

If you have no preference, GPT Image 2 for the frame → Seedance 2.0 for the animation. (See our Seedance vs Kling vs Veo breakdown.)

Step 5 — Generate and Iterate

First generations rarely nail it. Common failures and fixes:

The graphic is there from the first frame (no build-on). Strengthen the clean open: "The first second is a CLEAN frame with NO lower-third; the graphic MUST start absent and animate on." Add "graphic appearing fully-formed at 0s" to the avoid list.

The locked graphic warps or its numbers change. Add: "Once locked, the lower-third is a static image overlay — identical position, text, and values in every frame after it assembles."

The face over-moves (a wide smile, a big head turn). Dial it down: "Only a subtle breath, one slow blink, and a slight 5–10° head settle. No wide smile forming, no big head turn."

It pushes into a tight close-up. Add: "Camera nearly locked off; hold broadcast distance the whole clip — the face stays well under 40% of the frame."

The text is garbled. Render the still frame (Part A) on an image model strong at text, and keep the name a short surname.

Budget 3–5 generations before a keeper. If you're 8+ deep, reread the prompt and hunt for anywhere the graphic isn't described as a timed overlay.

Step 6 — Post It

The generation is half the work. How you post decides whether it gets 200 views or 2 million.

Caption framing. Post it like a real intro — "they finally put me in the starting lineup 😤" or "number eleven, checking in." That files it as sports content, not an AI demo. (Platforms increasingly require AI-disclosure labels — comply, but your caption copy still matters within those rules.)

Native 16:9, then crop. The broadcast look reads best in 16:9; generate wide, then center-crop to 9:16 for TikTok/Reels if you need vertical.

Timing. Post during a live game window — sports content peaks when real games are on.

Don't over-edit. No extra text overlays, no music bed on top. The clip already has its own graphic and PA audio; piling on gives away that you're trying to make it go viral.

Common Mistakes That Tank Your Video

1. The graphic never builds on. If it's static from frame one, you've lost the entire hook. Force the clean open.
2. Over-animating the face. The exact thing that got ESPN's real portraits roasted. Subtle breath and one blink, nothing more.
3. A locked graphic that keeps moving. After it assembles, it must freeze. Drifting or re-rendering numbers is an instant tell.
4. A long full name. Shrinks the type and garbles. Use a short surname.
5. A tight beauty close-up. Broadcast intros hold distance with the crowd visible. A glam crop kills the realism.
6. A real network or team logo. Rendered logos trip copyright moderation and break the de-branded look. Keep colors generic and the corner bug abstract.
7. A recognizable celebrity face. The model fights identity lock and viewers instantly clock it.

Window of Opportunity

Broadcast-realism formats have been the throughline of 2026, and the ESPN AI-portrait moment handed the player-intro version a wave of free attention. Trends like this have weeks, not months, before saturation. If a broadcast intro with your name on it is the goal and the two-part prompt above sounds like a chore, the Starting Lineup template at the top of this guide is this exact workflow in one tap.

Frequently Asked Questions

How do I make an ESPN-style AI video of myself? Render a finished broadcast "starting lineup" frame (you under intro spotlights with a lower-third showing your name, number, position, and a stat line), then write a video prompt that opens clean and animates the lower-third on before locking it while you stay subtly alive. Generate on Seedance 2.0, keep it de-branded, and post it like a real intro. Or use the Starrd Starting Lineup template to skip the prompt work.

What is the ESPN AI portrait trend? In June 2026, ESPN aired AI-generated "living portraits" of NBA legends during the Finals, got roasted as "AI slop," and pulled them after one game. The creator flip is putting yourself on the broadcast on purpose — a personalized player introduction with your own name and stats.

How do you get the broadcast graphic to animate in? Describe the lower-third as a timed 2D graphics overlay: tell the model the reference frame is the finished end state, open on a clean frame, build the graphic on in sequence over the first couple of seconds, then lock it to match the reference. Seedance 2.0 renders the animation and keeps the text crisp.

What AI model is best for the ESPN broadcast video? Seedance 2.0 for the animation (it can open clean and build a graphic on cue, with legible text and audio), and GPT Image 2 for the still end-state frame (strongest at dense on-screen text).

How is this different from a courtside or fan-cam video? A fan cam catches you as a spectator with no name. The player-intro version makes you the athlete being introduced, with a personalized name-and-stats lower-third that animates on. The named, animated chyron is the whole hook.

Why does my ESPN AI video look fake? Over-animating the face, a tight close-up instead of broadcast distance, a graphic that warps or changes numbers, or a real network/team logo. Keep the person still, the graphic locked after it builds on, and everything de-branded.

Do I need to disclose that the video is AI-generated? Yes — TikTok, Instagram, and YouTube require AI-generated content to be labelled. Keep it yourself or friends with consent, and never render a real network's logo or imply a real broadcast endorsed you.

Can I make it for a different sport or change my stats? Yes. Director's notes customize the name, number, position, stat values, sport, team color, and vibe — the animated-lower-third format stays the same.

Related Reading

• How to Make a Courtside NBA AI Video — the celebrity-row fan-cam cousin of this format
• How to Make the AI World Cup Fan Cam Video — the broadcast catches you in the stands
• How to Make the AI Soccer Poster — and Animate It Into a Lineup Reveal — the soccer campaign-poster version of a lineup graphic
• Seedance 2.0 Prompt Guide — the full framework for prompts that don't look like AI video
• Seedance vs Kling vs Veo — which model to pick and why